[pacman-dev] Checking free space before transaction

[Dan McGee]

On Thu, Nov 11, 2010 at 9:05 AM, Allan McRae <allan at archlinux.org> wrote:
> On 12/11/10 00:49, Nezmer wrote:
>>
>> On Thu, Nov 11, 2010 at 04:46:44PM +0200, Nezmer wrote:
>>>
>>> On Thu, Nov 11, 2010 at 09:09:27AM +1000, Allan McRae wrote:
>>>>
>>>> On 11/11/10 07:29, Nezmer wrote:
>>>>>
>>>>> But with those changes pacman did not like my ZFS pool.
>>>>
>>>>  From a bit of google, it looks like the statvfs call is probably
>>>> returning EOVERFLOW on that so we can probably just assume LOTS of
>>>> space and move on in those conditions...
>>>>
>>>> Allan
>>>>
>>>>
>>>
>>> That's over-simplification. This is the quick patch I needed to make
>>> this even compile.
>>
>>> diff --git a/lib/libalpm/diskspace.c b/lib/libalpm/diskspace.c
>>> index 51c25ed..f382d3c 100644
>>> --- a/lib/libalpm/diskspace.c
>>> +++ b/lib/libalpm/diskspace.c
>>> @@ -80,7 +80,7 @@ static alpm_list_t *mount_point_list()
>>>        endmntent(fp);
>>>  #elif defined HAVE_GETMNTINFO
>>>        int entries;
>>> -       struct statvfs *fsp;
>>> +       struct statfs *fsp;
>>>
>>>        entries = getmntinfo(&fsp, MNT_NOWAIT);
>>>
>>
>> from /usr/include/sys/mount.h:
>> int getmntinfo(struct statfs **, int);
>
> And here comes the joys of various BSDs doing whatever they feel like...   I
> have programmed based on the NetBSD man page which does use a statvfs
> struct...

This seems to explain it well:
http://stackoverflow.com/questions/1653163/difference-between-statvfs-and-statfs-system-calls

FreeBSD also seems to define the function and type elsewhere, so we
might just need to be smarter about what header we include:
http://www.cygwin.com/ml/cygwin-developers/2003-12/msg00020.html

So we shouldn't be using statfs() if possible. What was the error,
Nezmer? Compile output would have been nice; is it that the type and
function aren't even defined? Can you grep /usr/include/ for us to see
if statvfs is defined but needs some sort of #define to turn it on?

-Dan